微型机与应用
微型機與應用
미형궤여응용
MICROCOMPUTER & ITS APPLICATIONS
2014年
7期
83-86,90
,共5页
梁凯%曹广忠%徐佳瑛%邱洪%周受钦
樑凱%曹廣忠%徐佳瑛%邱洪%週受欽
량개%조엄충%서가영%구홍%주수흠
双 DSP 结构%CMOS 数字摄像头%手势识别%红外遥控
雙 DSP 結構%CMOS 數字攝像頭%手勢識彆%紅外遙控
쌍 DSP 결구%CMOS 수자섭상두%수세식별%홍외요공
dual-DSP structure%CMOS digital camera%gesture recognition%infrared telecontrol
为了在红外遥控系统中实现手势交互,设计了一种基于双 DSP 结构的具有动态手势视觉识别功能的红外遥控系统。针对手势识别图像算法的复杂性,系统采用了一种流水线式的动态图像处理模式并提出了一种基于双端口 RAM 通信的双 DSP 图像处理硬件系统结构。结构前级负责实时采集图像数据及手势图像预处理,后级结合以帧差法为核心的手势识别算法负责动作识别处理,从而节省单 DSP 需要的等待时间,手势图像处理任务同步进行,加快动态图像处理的速度。另外,双DSP 结构使得系统资源得到有效分配,复杂处理算法同步运行,再配合自学习红外遥控模块实现红外遥控功能。实验结果表明,系统一次手势动作的最短识别时间为1.5 s,动作识别的准确率达到93.75%,具有良好的稳定性和可操作性。
為瞭在紅外遙控繫統中實現手勢交互,設計瞭一種基于雙 DSP 結構的具有動態手勢視覺識彆功能的紅外遙控繫統。針對手勢識彆圖像算法的複雜性,繫統採用瞭一種流水線式的動態圖像處理模式併提齣瞭一種基于雙耑口 RAM 通信的雙 DSP 圖像處理硬件繫統結構。結構前級負責實時採集圖像數據及手勢圖像預處理,後級結閤以幀差法為覈心的手勢識彆算法負責動作識彆處理,從而節省單 DSP 需要的等待時間,手勢圖像處理任務同步進行,加快動態圖像處理的速度。另外,雙DSP 結構使得繫統資源得到有效分配,複雜處理算法同步運行,再配閤自學習紅外遙控模塊實現紅外遙控功能。實驗結果錶明,繫統一次手勢動作的最短識彆時間為1.5 s,動作識彆的準確率達到93.75%,具有良好的穩定性和可操作性。
위료재홍외요공계통중실현수세교호,설계료일충기우쌍 DSP 결구적구유동태수세시각식별공능적홍외요공계통。침대수세식별도상산법적복잡성,계통채용료일충류수선식적동태도상처리모식병제출료일충기우쌍단구 RAM 통신적쌍 DSP 도상처리경건계통결구。결구전급부책실시채집도상수거급수세도상예처리,후급결합이정차법위핵심적수세식별산법부책동작식별처리,종이절성단 DSP 수요적등대시간,수세도상처리임무동보진행,가쾌동태도상처리적속도。령외,쌍DSP 결구사득계통자원득도유효분배,복잡처리산법동보운행,재배합자학습홍외요공모괴실현홍외요공공능。실험결과표명,계통일차수세동작적최단식별시간위1.5 s,동작식별적준학솔체도93.75%,구유량호적은정성화가조작성。
In order to realize an infrared telecontrol system on user′s gesture interaction, a dual DSP-based system is designed. Due to the complexity of image algorithm on gesture recognition, a line-type image processing mode is used in the system and realized by using a dual-DSP processing platform which based on dual port RAM communication. The first stage of the processing mode is responsible for the real-time data acquisition and image preprocessing while the second stage combined with gesture image algorithm of frame difference and self-learning infrared telecontrol module is responsible for action recognition and telecontrol. Resources are efficiently allocated by Dual DSP structure so that waiting time needed by single DSP is saved and Image processing tasks are processed synchronously. Experimental results show that the shortest recognition time of system is 1.5 second as well as action recognition accuracy reaches 93.75%. The system has good stability and maneuverability.